A Multi-resource Selection Scheme for Virtual Machine Consolidation in Cloud Data Centers

Abstract

Resources used in a cloud data center could be spread across a large number of servers that are not fully utilized. This situation results in significant operational costs which are directly related to the power consumption of active servers. Virtual machine migration enables reducing the number of active servers by consolidating the load on a limited amount of nodes. Several schemes have actually been proposed to consolidate virtual machines on the minimum number of physical servers in order to reduce power consumption. However, most of the existing solutions only consider a limited tradeoff among multiple types of resources, thus resulting in unnecessarily activated physical servers. This article proposes a multi-resource selection (MRS) scheme for consolidating virtual machines in cloud data centers. With MRS, each physical server is first characterized in terms of multiple types of resources and then classified through its overall resource utilization. Based on the MRS scheme, a balanced multiple-resource utilization algorithm is also used to spread the load across different types of resources while consolidating virtual machines. The proposed solution is evaluated through simulations on both synthetic and real-world workloads. Experimental results show that the proposed approach outperforms several existing schemes in terms of the number of active physical servers and the utilization of multiple resources.

Virtual Machine Consolidation with Multiple Usage Prediction for Energy-Efficient Cloud Data Centers

“…In particular, VMCUP-M consolidates VMs according to the usage of multiple resources and a tunable horizon to predict future utilization. This article extends our previous work [18,19] as follows. We first present an efficient multiple usage prediction (MUP) approach to estimate the long-term utilization of each resource type based on the local history of the considered servers.…”

An algorithm for network and data-aware placement of multi-tier applications in cloud data centers

“…Since this approach does not consider inter-VM network communication patterns, such placement decisions can eventually result in locating VMs with high mutual network traffic in long distant servers, such as servers locating across the network edges. Several other VM placement works focusing on non-network objectives can be found in (Wu and Ishikawa, 2015), (Farahnakian et al, 2015), (Nguyen et al, 2014), (Corradi et al, 2014), and (Alboaneen et al, 2014). With a network-focused analysis, it can be concluded that research works such as the above ones considered single-tier applications and VM clusters without consideration of mutual network communication within the application components or VMs.…”

Virtual Machine Consolidation with Usage Prediction for Energy-Efficient Cloud Data Centers

“…Consolidation of VMs was considered in our earlier work [14] for reallocating the VMs to already-active servers. A multi-resource selection (MRS) scheme was proposed based on the current resource utilization and used as allocation criteria when determining a server is overloaded and underutilized.…”

“…An efficient VM consolidation algorithm was also proposed for spreading the load across different types of resources while consolidating VMs. With respect to [14], we predict the short-term future resource utilization based on past statistics of the considered servers. We use the current and predicted usage metrics as a reliable characterization of overloaded and underloaded servers.…”

“…The underloaded server migration procedure starts when no host is overutilized and the host with the minimum utilization is considered [6] (lines [14][15][16]. Recall that our model defines a cold server when its predicted resource usage is below the current utilization.…”